Results of the search

The search was conducted on the 12th November 2014.

The results of the search are presented in Figure 1. The search strategy identified more than 6000 records of which 31 were evaluated in full text to determine eligibility.

Figure 1

---

---

Study flow diagram.

There were 11 RCTs (Blasco 2012; Buchbinder 2009; Chen 2014; Dohm 2014; Farrokhi 2011; Kallmes 2009; Klazen 2010; Liu 2010; Rousing 2009; Vogl 2013; Voormolen 2007) and one quasi‐randomised trial (Endres 2012) that met inclusion criteria for this review. Six RCTs were registered in a trial registry (Blasco 2012; Buchbinder 2009; Dohm 2014; Farrokhi 2011; Kallmes 2009; Klazen 2010), although one was registered retrospectively (Farrokhi 2011). One RCT was stated to be registered on clinicaltrials.gov (NCT00576546) (Vogl 2013), but this could not be verified. No trial registration was found for the other five trials (Chen 2014; Endres 2012; Liu 2010; Rousing 2009; Voormolen 2007).

Three studies were excluded after full‐text assessment (Gilula 2013; Huang 2014; Yi 2014; ) see table of Characteristics of excluded studies).

An additional two papers were published protocols for RCTs (Longo 2010; (Firanescu 2011). Further details are unknown for one trial as we could not find any registration details (Longo 2010) (see table of Characteristics of ongoing studies), while the other one is reported to have been completed on clinicaltrials.gov as of 19 Nov 2014, but the results are not yet reported (see Characteristics of studies awaiting classification). An additional trial that appears to be unregistered has only been published as a conference abstract (Hao 2014) (also see Characteristics of studies awaiting classification).

Based upon a search of trial registries, we identified an additional five registered RCTs that are reported to have been completed but we were unable to find published results (Dolin 2003; Evans 2006; Laredo JD (OSTEO‐6); Laredo JD (STIC2); Sorensen 2005) (see table of Characteristics of studies awaiting classification). Another five registered RCTs are still recruiting participants (Carli D ('VERTOS V'); Clark W; Hansen EJ ('VOPE'); Nieuwenhuijse 2012; Sun 2012) and one registered RCT that is due to commence recruitment in 2015 (Zhao 2014) was also identified (see table of Characteristics of ongoing studies).

One additional registered trial is reported to have been suspended (Nakstad 2008) prior to completion (also see table of Characteristics of studies awaiting classification).

In summary, 11 RCTs and one quasi‐randomised trial have been included in this review, three studies were excluded, eight trials are awaiting classification, and seven trials are ongoing (including one that has not yet commenced).

Detailed descriptions of all unpublished trials that are either completed, suspended or ongoing are provided in either the table of Characteristics of studies awaiting classification or table of Characteristics of ongoing studies and a summary of all unpublished trials is provided in Table 1.

Included studies

A full description of all included trials is provided in the table of Characteristics of included studies and a summary of trial and participant characteristics is provided in Table 2.

Excluded studies

One controlled trial was excluded because participants were not assigned treatment at random but rather the authors stated that a surgeon at the outpatient ward blindly chose one of three different treatment modalities (vertebroplasty, kyphoplasty or usual care) to ensure similar pre‐treatment age, symptoms, grade and level of spinal diseases among the patients (Yi 2014). In addition, the vertebroplasty and kyphoplasty groups were combined in the data analysis and separate data for the vertebroplasty group were not provided.

Another two trials were excluded as vertebroplasty was administered to both treatment groups, with a different cement type in each group (Gilula 2013; Huang 2014).

Allocation

Two trials described adequate sequence generation and allocation concealment and were assessed as being at low risk of selection bias (Buchbinder 2009; Kallmes 2009). Dohm 2014 reported that randomisation was prepared by computer using a dynamic minimisation technique stratified by the number of prevalent vertebral fractures, aetiology and study centre, but treatment allocation was not concealed. Stratification by aetiology was unexplained as the selection criteria indicated that participants were included on the basis of osteoporotic fractures while fractures due to cancer and high‐energy trauma were excluded.

Two additional trials were also assessed as being at low risk of selection bias although allocation concealment was not explicitly reported (Farrokhi 2011; Klazen 2010). Farrokhi 2011 reported that the treatment assignment was kept in sealed envelopes. It is not clearly reported who prepared and opened the envelopes, but it is likely that allocation was concealed as they reported that neither the neurosurgeon (performing vertebroplasty) nor the physician (administering usual care) knew about the other study group and had no role in allocation. Klazen 2010 reported that an independent telephone operator allocated participants by telephone, therefore the allocation was likely concealed from the investigators.

Both Blasco 2012 and Liu 2010 reported that they prepared a computer‐generated random list but no information is provided regarding concealment of treatment allocation. Neither Chen 2014 nor Vogl 2013 described how randomisation was achieved or whether or not treatment allocation was concealed. In Rousing 2009, sealed envelopes containing the treatment allocation were prepared beforehand by the investigating surgeon and 'sorted randomly' and type of treatment was unknown to the patient and the investigators until after the patient had provided written consent. In Voormolen 2007, the patients were randomised in two groups by an independent central operator but no further information is provided regarding concealment of treatment allocation. Endres 2012 was assessed as being at high risk of selection bias with respect to random sequence generation as participants were reported to have been distributed quasi‐randomly into three groups and the method was not reported. It was judged to be at unclear risk of selection bias as the single investigator was not blinded to treatment allocation although participants were reported to be blinded and all procedures were performed under general anaesthesia.

Blinding

Two trials were judged to be at low risk of performance and detection bias for all clinical outcomes as they blinded participants and all study personnel other than the person performing the intervention (Buchbinder 2009; Kallmes 2009). However after the one month follow‐up, Kallmes 2009 was considered to be at high risk of performance bias because more participants in the placebo group crossed over (27/63, 36%) compared with the vertebroplasty group (8/68, 12%) by the three‐month follow‐up. Buchbinder 2009 was judged to be at high risk of bias for the one‐ and two‐year assessment of radiographically apparent incident fractures as it was not possible to blind radiologists to treatment allocation due to the opacity of the cement.

Seven trials were judged to be at high risk of performance and detection bias as participants and study personnel were aware of the treatment received (Blasco 2012; Chen 2014; Dohm 2014; Farrokhi 2011; Klazen 2010; Rousing 2009; Voormolen 2007). Liu 2010 was judged to be at unclear risk of detection bias for participant‐reported endpoints as it was unclear whether or not participants were blinded to treatment allocation. It was also judged to be at uncertain risk of bias for radiographic outcomes of vertebral height and kyphotic angle as while these were measured by technicians who were blinded to treatment allocation, it was not clear how variability of assessment was 'controlled via inter‐ and intra‐observer comparisons' (these outcomes were not reported in this review).

Vogl 2013 was judged to be at unclear risk of performance bias and low risk of detection bias for participant‐reported outcomes as participants were blinded to treatment allocation. However it was judged to be at high risk of detection bias for investigator‐reported outcomes as investigators were aware of treatment allocation. Endres 2012 was assessed as being at unclear risk of performance bias as participants, but not the single investigator, were blinded to treatment allocation, low risk of bias for self‐assessed outcomes as participants were blinded and another orthopaedic surgeon not involved in the primary surgery performed the final follow‐up, and high risk of bias for investigator‐reported outcomes as radiologic outcomes were analysed by the unblinded orthopaedic surgeon who performed all procedures as well as another radiologist (status of radiologist with respect to blinding not reported).

Incomplete outcome data

Three trials were assessed as at low risk for attrition bias (Buchbinder 2009; Farrokhi 2011; Kallmes 2009) while the others were either considered to be at unclear (Blasco 2012; Dohm 2014; Endres 2012; Klazen 2010; Liu 2010; Voormolen 2007) or high risk (Chen 2014; Rousing 2009; Vogl 2013).

Buchbinder 2009 had small and equal loss to follow‐up across treatment groups for shorter‐term benefit and safety outcomes although loss to follow‐up was greater when considering longer‐term outcomes. At two years, 29/38 (76%) and 28/40 (70%) had completed follow‐up in the vertebroplasty and sham groups respectively.

Kallmes 2009 had small and balanced loss to follow‐up across treatment groups up to one month.

Blasco 2012 was at unclear risk for attrition bias because while the proportion lost to follow‐up at 12 months was similar between groups (17/64 (27%) from the vertebroplasty group and 13/61 (21%) from the usual care group), the authors reported that the losses may not have been random, but related to worse pain in the usual care group.

Chen 2014 was judged to be at high risk of attrition bias as they performed a completers' analysis excluded 7/50 participants allocated to receive conservative care on the basis that four refused conservative treatment and decided to have vertebroplasty at the three‐month follow‐up and an additional three were lost to follow‐up. However while they stated that four participants in the vertebroplasty group were lost to follow‐up they appeared to include all 46 participants allocated to receive vertebroplasty in the analysis to 12 months.

Dohm 2014 had small and equal loss to follow‐up across treatment groups for shorter‐term efficacy and safety outcomes. However for the primary endpoints of new radiographic vertebral fracture at 12 and 24 months, loss to follow‐up was much greater. At 12 months 130/190 (68%) and 143/191 (75%) had completed follow‐up in the vertebroplasty and kyphoplasty groups respectively, while at 24 months complete follow‐up was 91/190 (48%) and 100/191 (52%) in the vertebroplasty and kyphoplasty groups respectively. While the reasons for loss to follow‐up were similar between groups, a higher proportion of those assigned to vertebroplasty (20/190; 11%) withdrew compared to the kyphoplasty group (11/191; 6%) and it is unclear if the reasons for withdrawal were systematically different.

In addition, seven participants who received vertebroplasty and four who received kyphoplasty underwent the alternate treatment for a subsequent vertebral fracture but the timing was not stated. An additional 70/88 (79.5%) participants with a new clinically recognised fracture underwent a subsequent vertebral augmentation during the trial ‐ it is implied that these participants received the same type of procedure as they had received as part of the trial. In all instances of spinal augmentation for a new vertebral fracture, the last observation before surgery was carried forward to later visits.

Endres 2012 was judged to be at unclear risk of attrition bias as data were unavailable for seven participants (two deaths and five participants who refused follow‐up although the treatment groups of these seven participants were not explicitly reported).

Klazen 2010 was judged to be at unclear risk of attrition bias as a greater number of participants completed one‐year follow‐up in the vertebroplasty group (86/101, 85%) compared with 77/101 (76%) in the usual care group and 15 (15%) participants in the usual care group received vertebroplasty.

Liu 2010 was judged to be at unclear risk for attrition bias because completeness of follow‐up was not explicitly reported.

Rousing 2009 was judged to be at high risk of attrition bias for several irregularities including failure to report baseline and follow‐up data for all participants.

Vogl 2013 was judged to be at high risk of attrition bias because of significant loss to follow‐up in both treatment arms (follow‐up complete at 12 months for 19 (68%) and 28 (57%) in the vertebroplasty and kyphoplasty groups respectively) and the reasons for missing data were not reported.

Voormolen 2007 was judged to be at unclear risk of bias as the treatment group of four participants excluded from the analysis due to refusal to complete two‐week follow‐up was not reported.

Selective reporting

Four trials were assessed as at low risk of reporting bias (Buchbinder 2009; Dohm 2014; Kallmes 2009; Klazen 2010), although Dohm 2014 included an additional outcome of time (in days) to new clinical vertebral fracture that was not pre‐specified.

Four trials were judged to be at unclear risk of bias because they did not appear to have been registered and did not publish a trial protocol (Chen 2014; Endres 2012; Liu 2010; Rousing 2009). In addition Chen 2014, only reported one‐day outcomes for mean pain.

Blasco 2012 was judged to be at unclear risk of reporting bias because adverse events were only reported for the vertebroplasty group and mean pain and quality of life and confidence intervals were reported graphically only. Voormolen 2007 was judged to be at unclear risk of bias because the number of participants with an incident clinical vertebral fracture was only reported for the vertebroplasty group and measures of variance were not reported for continuous outcomes.

Farrokhi 2011 was judged to be at unclear risk of reporting bias as it was unclear if any additional outcomes were measured and not reported.

Vogl 2013 was judged to be at high risk of report bias because in a published congress abstract of the same trial it was reported that pain intensity on a visual analogue scale (VAS) and disability assessed by the Oswestry Disability Index (ODI) were measured but only baseline pain was presented in the published paper. In addition the trial registration information could not be located.

Other potential sources of bias

No other sources of bias were detected for four trials (Blasco 2012; Buchbinder 2009; Kallmes 2009; Liu 2010).

Chen 2014 did not report on the number of 'prophylactic' vertebroplasties that were performed in the vertebroplasty group and did not specify a source of funding.

Dohm 2014 was sponsored by a device company which also contributed to study design, data monitoring, statistical analysis and reporting of results including manuscript authorship, paid for independent core laboratory and data safety‐monitoring board services, and terminated the study early.

Although Endres 2012 reported that there were no significant differences in baseline characteristics or planned vertebral treatment levels between treatment groups at baseline, participants in the vertebroplasty group appeared to be older on average than participants in the two other groups (71.3 versus 63.3 and 67.1 years in the balloon and shield vertebroplasty groups respectively). In addition, participants in the kyphoplasty groups also appeared to have worse pain and disability scores at baseline compared to the vertebroplasty group (vertebroplasty: 78.2 and 68.2; balloon kyphoplasty: 90.0 and 77.0; and shield kyphoplasty 88.16 and 75.7 respectively). They also did not state whether BioMedEs had any role in the study other than funding translation and copyediting.

Farrokhi 2011 noted that the trial was partially funded by Apadana Tajhizgostar Co., a distributor of medical devices, but its exact role in the trial was not explicitly reported.

Quality of life and disability were worse at baseline in the vertebroplasty group in Klazen 2010, which may have biased the results in favour of the vertebroplasty group. In addition, the role of COOK Medical (Bloomington, IN, USA), which provided an unrestricted grant, is not explicitly reported.

In the trial by Rousing 2009, baseline pain was higher in the usual care group (8.8 versus 7.5) and it was only measured in 17/24 and 19/25 participants in the usual care and vertebroplasty‐treated groups respectively. In addition, participants receiving usual care were hospitalised for longer (11.7 days versus 7.6 days); it is unclear if more pain medication and physiotherapy was offered, and how this would affect outcomes.

Vogl 2013 reported that Soteira Inc. (Natick, MA) funded the trial and provided the Cement Directed Kyphoplasty Systems but whether or not it had any other role in the trial was not explicitly reported.

In Voormolen 2007, eight participants withdrew after randomisation as they were not assigned to their preferred treatment (two in the vertebroplasty group and six in the usual care group). In addition, no specific source of funding was reported.

Benefits

1. Vertebroplasty versus placebo (sham)

The two placebo‐controlled trials were judged to be clinically homogenous with respect to baseline participant characteristics (see Table 2), allowing data to be pooled. The major outcomes for the primary comparison of vertebroplasty versus placebo (sham) is shown in summary of findings Table for the main comparison.

No between‐group differences in outcome were observed for any benefit outcome, with statistical heterogeneity considered not to be important (i.e. I²≤ 2%) for all pooled analyses, except for the one to two week pooled analysis for disability where the I² was 33% indicating the presence of moderate statistical heterogeneity.

There was no significant between‐group differences with respect to mean pain at one to two weeks (205 participants) or one month (201 participants) based upon pooled data from the two placebo‐controlled trials (mean difference (MD) 0.12 [95% confidence interval (CI) ‐0.65 to 0.88] and MD ‐0.66 [95% CI ‐1.48 to 0.15] at one month respectively) (Analysis 1.1). At one month, mean pain was five points on a 0 to 10 scale with placebo and 0.7 points lower (1.5 lower to 0.15 higher) with vertebroplasty, an absolute pain reduction of 7% (15% better to 1.5% worse) and relative reduction of 10% (21% better to 2% worse) (summary of findings Table for the main comparison).

Based upon data from one trial (Buchbinder 2009), no between‐group differences in pain were observed up to 24 months follow‐up. There was also no between‐group differences in the proportion of participants who improved from baseline by 2.5 units or 30% or more at one month based upon data from two trials (61/105 versus 46/101 respectively, (risk ratio (RR) 1.27 [95% CI 0.97 to 1.66]) (Analysis 1.2).

There were no significant between‐group differences with respect to disability (both measured with the RMDQ [0 to 23 scale]) at one to two weeks (190 participants) or one month (187 participants) based upon pooled data from the two placebo‐controlled trials (MD 0.82 [95% CI ‐1.13 to 2.78] and MD ‐1.09 [95% CI ‐2.94 to 0.76] respectively)(Analysis 1.3). At one month, the mean RDMQ was 13.6 points in the placebo group and 1.1 points lower (2.94 lower to 0.76 higher) in the vertebroplasty group, an absolute improvement in disability of 4.8% (12.8% better to 3.3% worse) and relative improvement of 6.3% (17.0% better to 4.4% worse) (summary of findings Table for the main comparison). There was also no between‐group differences in RMDQ scores up to 24 months based upon data from one trial (Buchbinder 2009).

There was no significant differences between vertebroplasty and placebo with respect to vertebral fracture and/or osteoporosis‐specific health‐related quality of life measured by the QUALEFFO up to 24 months based upon data from one trial (73 participants) (Analysis 1.4), or in overall quality of life (measured with the EQ5D) based upon data from two trials (187 participants)(MD 0.05 [95% CI ‐0.01 to 0.11] at one month (Analysis 1.5), or up to 24 months based upon data from a single trial (Buchbinder 2009).

One trial found no between‐group difference in treatment success up to 24 months (Analysis 1.6).

2. Vertebroplasty versus usual care

The six trials that compared vertebroplasty with usual care included participants with similar levels of baseline pain and disability and gender distribution was also similar across trials (see Table 2). Although one trial included younger participants (Chen 2014), and the duration of symptoms varied from one week to six months, we judged that the trials were sufficiently clinically homogenous to allow data to be pooled. For analyses including Farrokhi 2011 we used SMD for pain as (Farrokhi 2011 used a 1 to 10 pain scale in comparison to all other trials that used a 0 to 10 pain scale), but back‐transformed the SMD to MD on a 0 to 10 scale by multiplying the SMD and 95% CIs by the standard deviation (SD) of pain at baseline from the control group of the Klazen 2010 trial (SD 1.6). For disability which included data for the RMDQ and ODI, we back‐transformed the SMD to MD on the RMDQ (0 to 24 scale) by multiplying the SMD and 95% CIs by the standard deviation (SD) of pain at baseline also from the control group of the Klazen 2010 trial (SD 4.2). In instances where analyses within a single data plot required a mix of MD and SMD analyses we have shown the SMD in the plots and present both the SMD and MD in the results. For clarity, we have indicated where the MD was back‐transformed from the SMD to MD (either because Farrokhi 2011 was included together with other trials in the same meta‐analysis for mean pain or because both RMDQ and ODI were included in the same meta‐analysis for disability).

Based upon data from up to five trials (520 participants), participants in the vertebroplasty group had greater improvement in mean pain compared with those in the usual care group at one to two weeks (five trials, 520 participants, SMD ‐1.07 [95% CI ‐2.01 to ‐0.14], back‐transformed MD ‐1.72 [95% CI ‐3.22 to ‐0.22]), one month (two trials, 277 participants, SMD ‐0.94 [95% CI ‐2.30 to 0.43], MD ‐1.75 [95% CI ‐2.92 to ‐0.58]), three months (five trials, 520 participants, SMD ‐0.90 [95% CI ‐1.59 to ‐0.21], back‐transformed MD ‐1.44 [95% CI ‐2.54 to ‐0.34]), six months (four trials, 466 participants, SMD ‐0.88 [95% CI ‐1.71 to ‐0.04], back‐transformed MD ‐1.41 [95% CI ‐2.74 to ‐0.06]) and 12 months (five trials, 505 participants, SMD ‐0.83 [95% CI ‐1.55 to ‐0.11], back‐transformed MD ‐1.33 [95% CI ‐2.48 to ‐0.18]) (Analysis 2.1). However there was considerable statistical heterogeneity across all pooled pain analyses with the I² varying between 92% and 95%. Removing single trials from each analysis did not appreciably alter the results. At 24 months there was no between‐group difference in mean pain based upon one trial (77 participants, SMD ‐5.65 [95% CI ‐6.67 to ‐4.63], MD ‐0.45 [95% CI ‐0.90 to 0.01]).

There was no between‐group difference in the proportion of participants who reported moderate or severe residual pain at 12 months (vertebroplasty: 36% and 19%; usual care 34% and 18%, respectively) in one trial (Blasco 2012).

Based upon data from up to four trials (396 participants), improvement in disability also favoured the vertebroplasty group at one to two weeks (four trials, 387 participants, SMD ‐2.04 [95% CI ‐3.57 to ‐0.50], back‐transformed MD ‐8.57 [95% CI ‐15.00 to ‐2.10]), three months (four trials, 396 participants, SMD ‐1.90 [95% CI ‐3.76 to ‐0.04], back‐transformed MD ‐7.98 [95% CI ‐15.79 to ‐0.17]) and 24 months (one trial, 77 participants, SMD SMD ‐5.65 [95% CI ‐6.67 to ‐4.63], MD ‐12.00 [95% CI ‐12.94 to ‐11.06], but not one month (two trials, 271 participants, SMD ‐0.94 [95 CI ‐2.30 to 0.43], MD ‐2.05 [95% CI ‐2.56 to ‐1.54], six months (three trials, 354 participants, SMD ‐1.69 [95% CI ‐3.55 to 0.17], back‐transformed MD ‐7.10 [95% CI ‐14.91 to 0.18]) or 12 months (four trials, 389 participants, SMD ‐1.26 [95% CI ‐2.61 to 0.08], back‐transformed MD ‐5.29 [95% CI ‐10.96 to 0.76]). Significant statistical heterogeneity was also present for all analyses (I² ranging from 96% to 98%).

There was no significant between‐group differences with respect to vertebral fracture or osteoporosis‐specific quality of life at any time point measured by the QUALEFFO, based upon data from up to three trials (one to two weeks: three trials, 341 participants, MD ‐4.01 [95% CI ‐11.77 to 3.75], one month: one trial, 182 participants, MD ‐4.21 [95% CI ‐8.86 to 0.44], three months: two trials, 308 participants, MD ‐1.61 [95% CI ‐8.10 to 4.88], six months: two trials, 308 participants, (MD ‐0.83 [95% CI ‐6.48 to 4.83], and 12 months: two trials, 308 participants, MD ‐0.46 [95% CI ‐5.09 to 4.18])(Analysis 2.3). Statistical heterogeneity varied from unimportant to considerable across different time points (I² varying between 22% and 75%).

Overall quality of life measured by the EQ‐5D marginally favoured the vertebroplasty group at one to two weeks (one trial, 183 participants, MD 0.08 [95% CI 0.00 to 0.15]), one month (one trial, 183 participants, MD 0.09 [95% CI 0.01 to 0.16]), and three months (two trials, 215 participants, MD 0.10 [95% CI 0.00 to 0.20]), but not six months (one trial, 183 participants, MD 0.07 [95% CI ‐0.02 to 0.15]) or 12 months (two trials, 215 participants, MD 0.07 [95% CI ‐0.00 to 0.14])(Analysis 2.4). Statistical heterogeneity was unimportant for the pooled analyses (I² 0% to 22%). Treatment success was not reported in any of the trials comparing vertebroplasty with usual care.

3. Vertebroplasty versus kyphoplasty

No efficacy data relevant to this review could be extracted from Vogl 2013. The aim of this trial was to compare leakage rates between treatment groups and only vertebral height and wedge angle were measured as efficacy outcomes. Only one of the four trials reported an a priori sample size calculation (Dohm 2014). However this trial was terminated after recruiting 404 of the planned sample size of 1234 participants.

Based upon one trial (100 participants) (Liu 2010), there was no between‐group difference in pain three days following treatment with either vertebroplasty or balloon kyphoplasty (three‐day pain (SD) 0 to 10 point VAS scale: 2.3 (0.5) and 2.6 (0.6) respectively, MD ‐0.30 [95% CI ‐0.08 to ‐0.52], data not shown). Based upon data from two trials (141 participants) (Endres 2012; Liu 2010), there was no between‐group difference with respect to mean pain at six months following treatment with either vertebroplasty or balloon kyphoplasty (MD ‐0.08 [95% CI ‐0.41 to 0.24] (Analysis 3.1). This was consistent with the observed lack of between‐group difference in pain at two weeks, one, three, 12 and 24 months in Dohm 2014 (Analysis 3.1).

Based upon one trial (404 participants)(Dohm 2014), there was no between‐group difference in degree of improvement in disability at one, three, 12 or 24 months (Analysis 3.2). For example, at one month (mean (SD) ODI (0 to 100 scale, lower scores indicate less disability): vertebroplasty 34.6 (17.6), kyphoplasty: 36.2 (19.8), MD ‐1.60 [95% CI ‐5.70 to 2.50]). Based upon one trial (66 participants) (Endres 2012), the degree of improvement in disability at six months favoured balloon kyphoplasty over vertebroplasty (mean (SD) ODI: balloon kyphoplasty: 43.1 (19.5), vertebroplasty: 53.1 (8.5), MD ‐10.00 [95% CI ‐0.71 to ‐19.29], data not shown)), however at baseline, participants in the balloon kyphoplasty were more disabled (mean (SD) baseline ODI: 77.0 (4.2) compared with 68.2 (5.7) in the vertebroplasty group). No difference between vertebroplasty and shield kyphoplasty was observed (mean (SD) baseline and six‐month ODI in the shield kyphoplasty group: 75.7 (9.1) and 56.1 (7.6) respectively; MD ‐3.00 [95% CI ‐8.05 to 2.05], data not shown).

Based upon one trial (404 participants) (Dohm 2014), there was also no between‐group difference in degree of improvement in overall quality of life at one, three, 12 or 24 months (Analysis 3.3). For example, at one month (mean (SD) EQ‐5D: vertebroplasty 0.71 (0.19), kyphoplasty: 0.7 (0.19), MD 0.01 [95% CI ‐0.03 to 0.50]).

Harms

1. Vertebroplasty versus placebo (sham) or usual care

2. Vertebroplasty versus kyphoplasty

Subgroup Analysis

Data from the two placebo‐controlled trials were available for subgroup analysis comparing participants with pain duration ≤ 6 weeks versus > 6 weeks. Duration of pain did not influence outcome. In particular no differences were observed for pain duration ≤ 6 weeks versus > 6 weeks with respect to pain at one to two weeks (Analysis 6.1) or one month (Analysis 6.2), disability at one to two weeks (Analysis 6.3) or one month (Analysis 6.4) or quality of life at one month (Analysis 6.5).

Sensitivity Analyses

Sensitivity analyses confirmed that results varied widely when open trials comparing vertebroplasty with usual care were included in the analysis. While there was no significant between‐group difference in pain at one to two weeks when the analysis included all trials combined (seven trials, 725 participants, SMD ‐0.43 [95% CI ‐0.91, 0.06], back‐transformed MD ‐0.69 [95% CI ‐1.46 to 0.10], I² 90%), the analysis favoured the vertebroplasty group for pain at one month (four trials, 478 participants, SMD ‐0.91 [95% CI ‐1.71, ‐0.12], I² 94% and MD ‐1.32 [95% CI ‐2.05 to ‐0.59], I² 74%) and three months (five trials, 593 participants, SMD ‐0.79 [95% CI ‐1.38 to ‐0.20], back‐transformed MD ‐1.26 [95% CI ‐2.21 to ‐0.32], I² 91%).

Analyses including all trials combined, disability favoured vertebroplasty over control at one month (MD ‐1.26 [95% CI ‐2.51, ‐0.00], four trials, 472 participants, I² 57%), but not at one to two weeks (MD ‐1.60 [95% CI ‐3.32, 0.12], five trials, 510 participants, I² = 81%) or three months (MD ‐1.64 [95% CI ‐3.52, 0.24], four trials, 344 participants, I² 74%).